clip-02-05.jl

Load Julia packages (libraries) needed for the snippets in chapter 0

using StatisticalRethinking, Optim
gr(size=(600,300));

Plots.GRBackend()

snippet 3.2

p_grid = range(0, step=0.001, stop=1)
prior = ones(length(p_grid))
likelihood = [pdf(Binomial(9, p), 6) for p in p_grid]
posterior = likelihood .* prior
posterior = posterior / sum(posterior)
samples = sample(p_grid, Weights(posterior), length(p_grid));
samples[1:5]

5-element Array{Float64,1}:
 0.646
 0.477
 0.888
 0.652
 0.64 

snippet 3.3

Draw 10000 samples from this posterior distribution

N = 10000
samples = sample(p_grid, Weights(posterior), N);

10000-element Array{Float64,1}:
 0.522
 0.761
 0.633
 0.554
 0.809
 0.643
 0.421
 0.806
 0.548
 0.515
 ⋮    
 0.496
 0.422
 0.619
 0.668
 0.366
 0.48 
 0.785
 0.73 
 0.549

In StatisticalRethinkingJulia samples will always be stored in an MCMCChains.Chains object.

chn = MCMCChains.Chains(reshape(samples, N, 1, 1), ["toss"]);

Object of type Chains, with data of type 10000×1×1 Array{Float64,3}

Iterations        = 1:10000
Thinning interval = 1
Chains            = 1
Samples per chain = 10000
parameters        = toss

parameters
      Mean   SD   Naive SE  MCSE     ESS  
toss 0.638 0.1384   0.0014 0.0016 7681.007

Describe the chain

describe(chn)

Iterations        = 1:10000
Thinning interval = 1
Chains            = 1
Samples per chain = 10000
parameters        = toss

Empirical Posterior Estimates
──────────────────────────────────────────
parameters
      Mean   SD   Naive SE  MCSE     ESS
toss 0.638 0.1384   0.0014 0.0016 7681.007

Quantiles
──────────────────────────────────────────
parameters
      2.5% 25.0% 50.0% 75.0% 97.5%
toss 0.355 0.544 0.646 0.741  0.88

Plot the chain

plot(chn)

snippet 3.4

Create a vector to hold the plots so we can later combine them

p = Vector{Plots.Plot{Plots.GRBackend}}(undef, 2)
p[1] = scatter(1:N, samples, markersize = 2, ylim=(0.0, 1.3), lab="Draws")